Large model deployment method and system based on multi-level cache mechanism

Abstract

The invention provides a large model deployment method and system based on a multi-level cache mechanism, and the method comprises the steps: building a first-level dynamic semantic cache, preferentially feeding back a reconstruction answer when a new request is matched, starting a second-level context association cache for a missed request, and starting a second-level context association cache for a missed request; the method comprises the following steps: extracting an implicit semantic trajectory vector and constructing a dynamic association graph to identify a potential context dependency relationship, realizing progressive answer splicing, if coverage is still not performed, activating a third-level intention decision cache, performing scene classification and state marking by utilizing a multi-level analysis structure, and finally performing classification and state marking by utilizing a multi-level analysis structure. Based on a knowledge topology network, a query instruction is generated, multi-modal response elements are extracted to generate a combined response, a system synchronously executes multi-level cache collaborative optimization, a cache strategy is adjusted according to user feedback, cache data are adaptively expanded and eliminated, and a multi-level cache is obtained. According to the technical scheme provided by the embodiment of the invention, the response speed, the context understanding capability and the personalized service level of large model deployment are remarkably improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of large model deployment, and in particular to a large model deployment method and system based on a multi-level cache mechanism. Background Art

[0002] In intelligent AI customer service Q&A scenarios, users expect to receive instant, accurate, and personalized answers. This requires the system to not only respond quickly to user inquiries, but also understand complex semantic information and contextual associations to provide coherent answers. In addition, as the complexity of the conversation increases, the system needs to handle contextual dependencies across sessions and dynamically adjust the response strategy based on different interaction modes. Therefore, the intelligent AI customer service system needs to have efficient processing capabilities, accurate semantic understanding, and a flexible multimodal response generation mechanism.

[0003] Currently, many intelligent customer service systems use large model-based methods to improve the quality and accuracy of question and answer. These methods usually include a single-level cache mechanism to store common question and answer pairs, as well as some simple context processing logic. In order to improve the response speed, some systems also use pre-trained models to directly match or retrieve answers. However, this method mainly focuses on the construction and maintenance of static knowledge bases, and has limited ability to capture and process dynamic semantic features.

[0004] Although existing intelligent customer service solutions can meet basic question-and-answer needs to a certain extent, they still have several significant shortcomings: due to the lack of an effective caching mechanism, especially caching for dynamic semantic features, existing systems are often unable to quickly process repeated or similar query requests, resulting in long response times; most systems find it difficult to effectively identify and utilize cross-session contextual information, resulting in answers provided in continuous dialogue scenarios that are not coherent or accurate enough, affecting user experience; traditional methods usually generate answers based on fixed templates or simple keyword matching, lack the ability to deeply understand the user's true intentions, and find it difficult to provide highly personalized and rich response content. Summary of the invention

[0005] The embodiments of the present application provide a large model deployment method and system based on a multi-level cache mechanism to solve the problem of lack of an effective cache mechanism in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a large model deployment method based on a multi-level cache mechanism, comprising the following steps: Establish a first-level dynamic semantic cache to capture the semantic features of online sessions in real time and encode them to generate unique identifiers. Count high-frequency semantic fragments based on dynamic access weights. When the overlap between the semantic identifier of a new request and the cached fragment meets the preset conditions, aggregate the discrete semantic fragments to reconstruct the complete response logic and give priority feedback. For missed requests, the second-level context association cache is started to extract the implicit semantic trajectory vector of the session, and a dynamic association graph is constructed in combination with the historical interaction path to identify potential context dependencies across sessions. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, the progressive answer splicing mechanism is triggered. When the second-level cache is not covered, the third-level intent decision cache is activated, the request is classified into scenarios and marked with status through a multi-level parsing structure, structured query instructions are generated based on the knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and multimodal response elements are extracted to generate a combined response; Synchronously perform multi-level cache collaborative optimization, dynamically adjust cache semantic coverage and matching priority according to large model response results and user feedback, and realize cache capacity adaptive expansion and elimination logic linkage based on session traffic characteristics; Among them, the semantic identifiers of the first-level cache are orthogonally complementary to the semantic trajectory vector space of the second-level cache, the association graph nodes of the second-level cache and the knowledge topology network of the third-level cache are mapped through the semantic bridging layer, and the update conditions of the caches at all levels are dynamically negatively correlated with the complexity of the session flow.

[0007] Optionally, the second-level context association cache is started for the missed request, the implicit semantic trajectory vector of the session is extracted, a dynamic association graph is constructed in combination with the historical interaction path, and potential context dependencies across sessions are identified. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, a progressive answer splicing mechanism is triggered, including: Extract the context encoding features of the continuous interaction sequence in the current session and generate an implicit semantic trajectory vector with temporal dependencies; The implicit semantic trajectory vector is aligned with the interaction path of the same user identifier or similar user group in the historical session library in time and space, and a dynamic association graph including semantic nodes and relationship edges is constructed, wherein the graph nodes store the semantic context fragments across sessions, and the relationship edges record the logical jump probability and timeliness weight between nodes; Performing bidirectional semantic propagation calculation on the dynamic association graph, activating adjacent nodes on the association path based on the semantic trajectory vector of the current request, and identifying multi-hop semantic chains that form logical coherence with the current session through a path confidence screening algorithm; When it is detected that there are at least two or more multi-hop semantic chains with time continuity and semantic Figure 1When a consistent associated node is found, the reconstruction engine of the cross-session answer fragments is triggered, and the discrete answer elements are dynamically sorted and connected according to the logical order and weight ratio between the nodes, generating a progressive spliced ​​response that is coherent with the current context.

[0008] Optionally, when the second-level cache is not covered, the third-level intent decision cache is activated, the request is classified into scenarios and marked in status through a multi-level parsing structure, a structured query instruction is generated based on a knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and a multimodal response element is extracted to generate a combined response, including: The session state awareness module captures the intent expression pattern and scenario feature parameters in the current interaction sequence, and generates a multidimensional classification vector containing domain labels and state variables. Mapping the multidimensional classification vector to a pre-constructed knowledge topology network, generating a structured query instruction with path constraints according to the domain association strength of each subgraph in the network, wherein the query instruction includes a search condition priority and a cross-domain fusion rule; Real-time monitoring of focus shifts and feedback signals in user interaction behaviors, dynamic calculation of the credibility attenuation factor of each node in the knowledge retrieval path, and adjustment of the scope and depth of cross-domain retrieval based on the path weight redistribution algorithm; According to the updated retrieval path, multimodal data units such as text, images, and operation instructions are extracted in parallel from heterogeneous data sources. After filtering the conflicting elements through the semantic consistency verification module, the multimodal units are embedded in the dynamically generated dialogue framework according to the preset response assembly template to form a combined response.

[0009] Optionally, the synchronous execution of multi-level cache collaborative optimization dynamically adjusts the cache semantic coverage and matching priority according to the large model response results and user feedback, and realizes cache capacity adaptive expansion and elimination logic linkage based on session traffic characteristics, including: Build a semantic traceability link to reversely decompose the final response generated by the large model into a contribution distribution graph of multi-level cache hit results, and calculate the semantic alignment error of each level of cache based on the dependency relationship between graph nodes; Collect user satisfaction scores and follow-up behavior data when the session ends, establish a cache effect evaluation matrix, and obtain the semantic coverage blind spots and matching rule deviations that need to be corrected for caches at all levels through matrix feature decomposition; Analyze the spatiotemporal distribution characteristics of session traffic in real time, build a cache load prediction model based on sliding windows, and dynamically calculate the capacity pressure coefficient of each level of cache and the migration trend of hot data; According to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, the semantic matching thresholds and storage partitioning strategies of caches at all levels are adjusted synchronously, and priority reordering and redundant data elimination of cross-layer caches are triggered based on hotspot migration trends.

[0010] Optionally, the real-time monitoring of the focus shift and feedback signals in the user interaction behavior, dynamically calculating the credibility attenuation factor of each node in the knowledge retrieval path, and adjusting the scope and depth of the cross-domain retrieval based on the path weight redistribution algorithm include: Based on the user interaction behavior sequence, the attention transfer matrix is ​​constructed, the current interaction focus and historical focus sequence are extracted through the sliding window mechanism, the focus transfer probability is calculated and the focus transfer vector is generated; Define the credibility attenuation factor in the knowledge topology network, calculate the node weight correction value based on user feedback signals, and update the node weight; Based on the updated node weights, a path weight redistribution algorithm is used to calculate the priority and depth limit of the cross-domain retrieval path; A multimodal retrieval instruction is generated according to the path weight coefficient and depth limit, driving the parallel extraction module to extract data units according to the weight ratio and input them into the semantic consistency verification module.

[0011] Optionally, according to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, the semantic matching threshold and storage partition strategy of caches at all levels are synchronously adjusted, and priority reordering and redundant data elimination of cross-layer caches are triggered based on the hotspot migration trend, including: Construct a cross-layer association model between the semantic alignment error matrix and the capacity pressure coefficient matrix, extract the collaborative optimization parameters of multi-level caches through the feature fusion algorithm, and generate a dynamic weight allocation map; Based on the dynamic weight allocation graph, a path optimization algorithm is used to calculate the semantic matching threshold adjustment amount of each level of cache, and a storage partition strategy update instruction is generated; Build a priority reordering queue based on the hotspot migration trend, and generate a cross-layer cache priority mapping table based on the temporal and spatial distribution characteristics of session traffic; The semantic matching threshold adjustment amount, the storage partition strategy update instruction and the cross-layer cache priority mapping table are input into the cache strategy execution engine, and the threshold update, partition reconstruction and redundant data elimination operations are triggered synchronously, and the adjustment results are fed back to the semantic traceability link for calibration.

[0012] Optionally, the real-time analysis of the spatiotemporal distribution characteristics of session traffic, the construction of a cache load prediction model based on a sliding window, and the dynamic calculation of the capacity pressure coefficient of each level of cache and the hotspot data migration trend include: Based on the sliding window mechanism, the session traffic time series is divided, the interaction frequency, semantic complexity and response delay parameters within the window are extracted, and the spatiotemporal feature vector is generated; The spatiotemporal feature vector is input into the hierarchical prediction model, the short-term traffic fluctuation characteristics are extracted through the time series convolution network, the cross-window traffic evolution law is captured by combining the long short-term memory network, and the capacity pressure coefficient of each level of cache is output; Based on the capacity pressure coefficient, a hotspot data identification engine is constructed, high-frequency access semantic fragments are screened through a dynamic threshold algorithm, and a hotspot migration path map is generated in combination with semantic topological relationships; The capacity pressure coefficient and the hotspot migration path map are input into the cache expansion decision module to trigger the dynamic reorganization of storage partitions and the redundant data elimination mechanism, and the load status mark of the semantic traceability link is synchronously updated.

[0013] In a second aspect, an embodiment of the present application provides a large model deployment system based on a multi-level cache mechanism, including: The capture module is used to establish the first-level dynamic semantic cache, capture the semantic features of online sessions in real time, encode and generate unique identifiers, count high-frequency semantic fragments based on dynamic access weights, and aggregate discrete semantic fragments to reconstruct the complete response logic and give priority feedback when the overlap between the semantic identifier of the new request and the cached fragment meets the preset conditions; The recognition module is used to start the second-level context association cache for missed requests, extract the implicit semantic trajectory vector of the session, build a dynamic association graph based on the historical interaction path, identify the potential context dependencies across sessions, and trigger the progressive answer splicing mechanism when continuous semantic nodes that form a logical link with the current request are detected in the association graph; The extraction module is used to activate the third-level intent decision cache when the second-level cache is not covered, classify the request scenarios and mark the status through a multi-level parsing structure, generate structured query instructions based on the knowledge topology network, dynamically adjust the knowledge retrieval path weight, and extract multimodal response elements to generate a combined response; The adjustment module is used to synchronously perform multi-level cache collaborative optimization, dynamically adjust the cache semantic coverage and matching priority according to the large model response results and user feedback, and realize the adaptive expansion of cache capacity and elimination logic linkage based on session traffic characteristics.

[0014] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a large model deployment method based on a multi-level cache mechanism as described in the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a large model deployment method based on a multi-level cache mechanism as described in the first aspect.

[0016] In the embodiment of the present application, a first-level dynamic semantic cache is established to capture the semantic feature encoding of the online session in real time to generate a unique identifier, and high-frequency semantic fragments are counted based on dynamic access weights. When the overlap between the semantic identifier of the new request and the cache fragment meets the preset conditions, the discrete semantic fragments are aggregated to reconstruct the complete response logic and give priority to feedback; the second-level context association cache is started for missed requests, the implicit semantic trajectory vector of the session is extracted, and a dynamic association graph is constructed in combination with the historical interaction path to identify potential context dependencies across sessions. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, a progressive answer splicing mechanism is triggered; when the second-level cache is not covered, the third-level intention decision cache is activated, the request is classified and the status is marked through a multi-level parsing structure, a structured query instruction is generated based on the knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and multimodal response elements are extracted to generate a combined response; The technical solution of this application has the following beneficial effects: Multi-level cache collaborative optimization is performed synchronously, and the cache semantic coverage and matching priority are dynamically adjusted according to the large model response results and user feedback. The adaptive expansion of cache capacity and the linkage of elimination logic are realized based on the session traffic characteristics. Among them, the semantic identifiers of the first-level cache and the semantic trajectory vector space of the second-level cache are orthogonally complementary, the association graph nodes of the second-level cache and the knowledge topology network of the third-level cache are mapped through the semantic bridging layer, and the cache update conditions at all levels are dynamically negatively correlated with the complexity of the session flow.

[0017] By establishing a first-level dynamic semantic cache, the semantic features of online sessions can be captured in real time and a unique identifier can be generated. When the semantic identifier of a new request meets the overlap requirements with the cached fragment, the discrete semantic fragments can be quickly aggregated to reconstruct the complete answer logic, achieving fast feedback, thereby significantly improving the response speed of the question-answering system. For requests that do not hit the first-level cache, the second-level context-associative cache is started to extract the implicit semantic trajectory vector of the session and construct a dynamic association graph to identify potential context dependencies across sessions and trigger a progressive answer splicing mechanism. This helps the system better understand and handle complex, continuous conversation scenarios, and provide more coherent and accurate answers. The third-level intent decision cache classifies requests by scenario and status through a multi-level parsing structure, and generates structured query instructions based on the knowledge topology network, which can more accurately capture the user's actual needs, extract multimodal response elements to generate combined responses, and further improve the question-answering system's intent parsing capabilities and the richness of answers. It simultaneously performs multi-level cache collaborative optimization and dynamically adjusts the cache strategy based on the large model response results and user feedback, which can not only effectively manage cache capacity and avoid unnecessary storage consumption, but also realize adaptive expansion and elimination logic linkage based on session traffic characteristics, ensuring the stable operation of the system under high concurrency conditions while improving user experience. The mutual complementation and dynamic adjustment mechanism between caches at all levels enable the system to flexibly respond to various complex user interaction modes, especially for tasks that need to be completed across multiple conversation rounds, and can provide a smoother and more natural communication experience.

[0018] Furthermore, the solution includes extracting the implicit semantic trajectory vector of the current session and aligning it with the historical interaction path in time and space to construct a dynamic association graph. The multi-hop semantic chain that forms logical coherence with the current request is identified through bidirectional semantic propagation calculation. When at least two temporally continuous and semantically coherent chains are detected, Figure 1 When the corresponding associated nodes are found, the reconstruction engine of the cross-conversation answer fragments is triggered, and the discrete answer elements are sorted and connected according to the logical order and weight ratio to generate a progressive spliced ​​response. The method described in the core solution can significantly enhance the ability of the intelligent AI customer service system to handle complex conversation situations. By effectively utilizing the current conversation and historical interaction data, the system can more accurately identify and understand the contextual dependencies across conversations, and provide more coherent and natural answers. This method not only improves the relevance and accuracy of the answers, but also enhances the user experience, allowing the system to maintain logical consistency in continuous conversations, effectively solving the challenges faced by traditional methods in handling complex and continuous conversations, and achieving a more intelligent and personalized customer service experience.

[0019] Furthermore, when the second-level cache does not cover the request, the process of activating the third-level intent decision cache. This process involves classifying the scenarios and marking the status of user requests through a multi-level parsing structure, and generating structured query instructions based on the constructed knowledge topology network. At the same time, the weight of the knowledge retrieval path is dynamically adjusted to optimize the search efficiency, and multimodal response elements such as text and images are extracted from heterogeneous data sources. After semantic consistency verification, these elements are integrated into a combined response according to the preset template. The ability of the intelligent AI customer service system to handle complex and diverse user requests is significantly improved. By accurately classifying the scenarios and marking the status of user requests and optimizing the retrieval path using the knowledge topology network, this method can not only quickly locate the most relevant answers, but also effectively integrate multiple types of response elements to provide rich and personalized answers. This method greatly enhances the accuracy of the system's intent parsing and the diversity of the response content, allowing the intelligent customer service to not only understand the user's direct needs, but also capture and respond to their potential intentions, thereby providing a more comprehensive and satisfactory customer service experience.

[0020] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flowchart of a large model deployment method based on a multi-level cache mechanism provided by the present application is shown; Figure 2 A schematic diagram of the structure of a large model deployment system based on a multi-level cache mechanism provided by the present application is shown; Figure 3 A schematic diagram of the structure of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0026] Figure 1 A flowchart of a large model deployment method based on a multi-level cache mechanism is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes: 101. Establish a first-level dynamic semantic cache, capture the semantic features of online sessions in real time, encode and generate unique identifiers, count high-frequency semantic fragments based on dynamic access weights, and aggregate discrete semantic fragments to reconstruct the complete response logic and give priority feedback when the overlap between the semantic identifier of a new request and the cached fragment meets the preset conditions; In this step, the first-level dynamic semantic cache includes the semantic feature encoding and unique identifier of the online session captured in real time. These data are used to quickly match new requests with the stored high-frequency semantic fragments to achieve efficient response to user queries.

[0027] In the embodiment of the present application, the system first establishes a dynamic semantic cache, generates semantic feature encoding by real-time analysis of user input in the intelligent AI customer service question and answer scenario, and creates a unique identifier for each conversation fragment. High-frequency semantic fragments based on access frequency statistics are stored first. When a new user request arrives, the system calculates the overlap between its semantic identifier and the fragment in the cache. If the preset conditions are met, the relevant semantic fragments are aggregated to reconstruct the complete response logic and quickly feedback to the user.

[0028] Suppose a user asks about the battery life of a certain mobile phone on the intelligent customer service platform. The system captures the semantic features of this query in real time and finds that it highly overlaps with a high-frequency semantic fragment in the cache. It then directly extracts the relevant information from the cache, combines it into a complete answer and immediately replies to the user.

[0029] 102. Start the second-level context association cache for missed requests, extract the implicit semantic trajectory vector of the session, build a dynamic association graph based on the historical interaction path, identify potential context dependencies across sessions, and trigger the progressive answer splicing mechanism when continuous semantic nodes that form a logical link with the current request are detected in the association graph; In this step, the second-level context association cache contains the session implicit semantic trajectory vector and the dynamic association graph constructed by the historical interaction path. These elements help identify potential context dependencies across sessions, thereby supporting more complex dialogue management.

[0030] In the embodiment of the present application, in the intelligent AI customer service question-and-answer scenario, for requests that cannot find answers in the first-level cache, the system starts the second-level cache mechanism, extracts the implicit semantic trajectory vector of the current session, and builds a dynamic association graph in combination with the historical interaction path. By identifying continuous semantic nodes related to the current request in the graph, the system triggers a progressive answer splicing mechanism to provide a more coherent answer.

[0031] Suppose a user mentions a question about a mobile phone model that he or she has consulted before in a conversation. The system analyzes the correlation between the current conversation and the historical interaction path, identifies the related context, and provides more personalized service suggestions based on this, such as recommending a mobile phone model and its features that are suitable for the user.

[0032] 103. When the second-level cache is not covered, the third-level intent decision cache is activated, the request is classified into scenarios and marked with status through a multi-level parsing structure, structured query instructions are generated based on the knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and multimodal response elements are extracted to generate a combined response; In this step, the third-level intent decision cache involves multi-level parsing structures, knowledge topology networks, and structured query instructions. These tools work together to parse the scene classification and state tags of user requests in order to accurately extract multimodal response elements and generate combined responses.

[0033] In the embodiment of the present application, in the intelligent AI customer service question and answer scenario, when the second-level cache cannot cover the request, the system activates the third-level cache, uses a multi-level parsing structure to analyze the request in detail, and uses the knowledge topology network to generate structured query instructions. Adjust the retrieval path weight according to the actual needs of the user, extract multimodal data units from different data sources, and integrate them into the final answer after consistency verification.

[0034] Suppose a user asks a multi-faceted question on the intelligent customer service platform, such as asking about the battery life, camera quality and price of a certain mobile phone. The system deeply analyzes the user's intention, uses the knowledge topology network to locate relevant information, and integrates data in various forms such as text and pictures to provide a comprehensive and accurate answer, including battery life test reports, camera samples and the latest market price information.

[0035] 104. Synchronously perform multi-level cache collaborative optimization, dynamically adjust cache semantic coverage and matching priority according to large model response results and user feedback, and realize cache capacity adaptive expansion and elimination logic linkage based on session traffic characteristics; Among them, the semantic identifiers of the first-level cache are orthogonally complementary to the semantic trajectory vector space of the second-level cache, the association graph nodes of the second-level cache and the knowledge topology network of the third-level cache are mapped through the semantic bridging layer, and the update conditions of the caches at all levels are dynamically negatively correlated with the complexity of the session flow.

[0036] In this step, the multi-level cache collaborative optimization mechanism covers functions such as cache semantic coverage, matching priority adjustment, adaptive expansion, and elimination logic linkage. These functions ensure the flexibility and efficiency of the system, enabling it to self-adjust according to actual usage.

[0037] In the embodiment of the present application, in the intelligent AI customer service Q&A scenario, the system synchronously performs multi-level cache collaborative optimization, continuously adjusts the cache strategy based on the large model response results and user feedback, and monitors the session traffic characteristics to achieve automatic capacity adjustment. In addition, the semantic bridge layer connects the caches at all levels to ensure effective collaboration between them.

[0038] Assume that with the arrival of the new product release season, the number of inquiries on the intelligent customer service platform has surged, especially information inquiries about new mobile phones. The system automatically expands the cache capacity and optimizes the matching rules to ensure that high-priority issues (such as new product features, time to market, etc.) are responded to in a timely manner, while maintaining the efficient operation of the overall system, reducing waiting time and improving user satisfaction.

[0039] In summary, steps 101 to 104 cover a full range of intelligent customer service solutions from quick response to complex context processing to personalized service provision, aiming to provide a flexible and efficient question-and-answer platform to meet the diverse consulting needs of users in intelligent AI customer service question-and-answer scenarios.

[0040] In order to further improve the processing capability of missed requests, in some embodiments, the second-level context association cache is started for the missed requests in step 102, the implicit semantic trajectory vector of the session is extracted, a dynamic association graph is constructed in combination with the historical interaction path, and potential context dependencies across sessions are identified. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, a progressive answer splicing mechanism is triggered, including: Extract the context encoding features of the continuous interaction sequence in the current session to generate an implicit semantic trajectory vector with temporal dependency; align the implicit semantic trajectory vector with the interaction path of the same user identifier or similar user group in the historical session library in time and space to construct a dynamic association graph containing semantic nodes and relationship edges, wherein the graph nodes store the semantic context fragments across sessions, and the relationship edges record the logical jump probability and timeliness weight between nodes; perform bidirectional semantic propagation calculation on the dynamic association graph, activate the adjacent nodes on the association path based on the semantic trajectory vector of the current request, and identify the multi-hop semantic chain that forms logical coherence with the current session through the path confidence screening algorithm; when it is detected that there are at least two or more semantic chains with temporal continuity and semantic coherence in the multi-hop semantic chain, the semantic chain is automatically identified. Figure 1 When a consistent associated node is found, the reconstruction engine of the cross-session answer fragments is triggered, and the discrete answer elements are dynamically sorted and connected according to the logical order and weight ratio between the nodes, generating a progressive spliced ​​response that is coherent with the current context.

[0041] In this embodiment, the second-level context association cache involves data structures such as implicit semantic trajectory vectors, dynamic association graphs, and multi-hop semantic chains. These data are used to capture and understand the user's interaction history, identify potential context dependencies, and generate coherent answers based on them.

[0042] In the embodiment of the present application, the context encoding features in the current session are firstly extracted to generate an implicit semantic trajectory vector; secondly, the vector is spatially and temporally aligned with the historical interaction path to construct a dynamic association graph; thirdly, a bidirectional semantic propagation calculation is performed to identify the multi-hop semantic chain related to the current request; finally, when a qualified association node is detected, the answer fragment reconstruction engine is triggered to generate a progressive splicing response according to the logical order and weight ratio; Here is a specific example: Suppose a user mentions a router configuration question that he had asked before when consulting about how to set up a home network. The system first extracts the relevant contextual features of the current session and generates an implicit semantic trajectory vector. Secondly, the vector is aligned with the interaction path of the same user in the historical session library to build a dynamic association graph containing multiple semantic nodes and their relationship edges. Thirdly, a two-way semantic propagation calculation is performed in the graph to find the multi-hop semantic chain related to the current query. Finally, if more than two continuous and consistent association nodes are found, the answer fragment reconstruction engine is triggered to combine relevant information in a logical order and weight ratio to provide a coherent and detailed answer to help the user successfully complete the home network setup. Through the above steps, not only the consistency and accuracy of the conversation are improved, but also the user experience is enhanced, enabling intelligent customer service to better understand and respond to complex needs.

[0043] In order to further improve the processing capability of complex user requests, in some embodiments, when the second-level cache is not covered in step 103, the third-level intent decision cache is activated, the request is classified into scenarios and marked in status through a multi-level parsing structure, a structured query instruction is generated based on a knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and a multimodal response element is extracted to generate a combined response, including: The intention expression mode and scenario feature parameters in the current interaction sequence are captured through the session state perception module, and a multidimensional classification vector containing domain labels and state variables is generated; the multidimensional classification vector is mapped to a pre-constructed knowledge topology network, and a structured query instruction with path constraints is generated according to the domain association strength of each subgraph in the network, and the query instruction includes retrieval condition priority and cross-domain fusion rules; the focus shift and feedback signals in user interaction behaviors are monitored in real time, the credibility attenuation factor of each node in the knowledge retrieval path is dynamically calculated, and the scope and depth of cross-domain retrieval are adjusted based on the path weight redistribution algorithm; according to the updated retrieval path, multimodal data units of text, image, and operation instructions are extracted from heterogeneous data sources in parallel, and after filtering the conflicting elements through the semantic consistency verification module, the multimodal units are embedded in the dynamically generated dialogue framework according to the preset response assembly template to form a combined response.

[0044] In this embodiment, the third-level intention decision cache involves data structures such as multi-dimensional classification vectors, knowledge topology networks, structured query instructions, and multi-modal data units. These data are used to accurately capture the user's intentions and scenario requirements, and generate personalized answer content accordingly.

[0045] In the embodiment of the present application, the session state perception module is first used to analyze the current interaction sequence, identify the intent expression mode and scene features, and generate a multi-dimensional classification vector; secondly, the vector is mapped to the knowledge topology network, and a structured query instruction is generated according to the domain association strength; thirdly, the user's interaction behavior and feedback are tracked in real time, and the node weights in the knowledge retrieval path are adjusted; finally, multimodal data is extracted from different data sources according to the optimized retrieval path, and integrated into the dialogue framework after consistency verification to provide a comprehensive answer; Here is a specific example: Suppose a user asks about how to solve the problem that his smart watch cannot be synchronized with the mobile phone. The system first analyzes the user's specific problem through the session state awareness module, determines his intention and scenario, and generates a multi-dimensional classification vector; secondly, the vector is mapped to the knowledge topology network, identifies the relevant technical fields, and generates structured query instructions with priority and cross-domain fusion rules; thirdly, the system monitors the user's interactive behavior in real time, such as whether the user's attention is shifted to a specific solution or new questions arise, and dynamically adjusts the node weights in the retrieval path; finally, based on the updated retrieval path, the system extracts relevant information from multiple data sources such as text guides, operation videos and troubleshooting steps, and after semantic consistency verification, integrates it into a detailed answer according to the preset template to help users gradually solve the synchronization problem between the watch and the mobile phone; through the above steps, not only the understanding and response efficiency of complex problems are improved, but also it is ensured that the solutions provided are both comprehensive and accurate, which greatly improves the user experience.

[0046] In order to further improve the accuracy of capturing and responding to user interaction behaviors, in certain embodiments, the real-time monitoring of focus shifts and feedback signals in user interaction behaviors, dynamically calculating the credibility attenuation factor of each node in the knowledge retrieval path, and adjusting the scope and depth of cross-domain retrieval based on the path weight redistribution algorithm include: Based on the user interaction behavior sequence, an attention transfer matrix is ​​constructed, and the current interaction focus and historical focus sequence are extracted through the sliding window mechanism. The focus transfer probability is calculated and the focus transfer vector is generated. The credibility attenuation factor is defined in the knowledge topology network, and the node weight correction value is calculated in combination with the user feedback signal to update the node weight. Based on the updated node weight, the path weight redistribution algorithm is used to calculate the priority and depth limit of the cross-domain retrieval path. According to the path weight coefficient and depth limit, a multimodal retrieval instruction is generated to drive the parallel extraction module to extract data units according to the weight ratio and input them into the semantic consistency verification module.

[0047] In this embodiment, concepts such as attention transfer matrix, focus transfer vector, credibility attenuation factor and path weight redistribution algorithm are involved. These data structures are used to dynamically analyze the user's interaction mode and interest changes, optimize the knowledge retrieval path, and ensure that the information provided is both relevant and accurate.

[0048] In the embodiment of the present application, firstly, an attention transfer matrix is ​​established based on the user's behavior sequence, the sliding window technology is used to identify the current and historical interaction focus, and the focus transfer probability is calculated to form a focus transfer vector; secondly, a credibility attenuation factor is set in the knowledge topology network, and the node weight is adjusted in combination with user feedback; thirdly, according to the updated node weight, the path weight redistribution algorithm is used to determine the priority and depth of the cross-domain retrieval path; finally, a multimodal retrieval instruction is generated based on the path weight and depth limit, and the parallel extraction module is guided to obtain data according to the weight ratio, and integrated into the response content after semantic consistency verification; The following is a specific example: suppose a user is looking for ways to solve the problem of his computer frequently restarting. The system first establishes an attention transfer matrix based on the user's interactive behavior sequence, identifies whether the user is mainly concerned about hardware failure or software problems through a sliding window mechanism, calculates the focus transfer probability and generates a focus transfer vector; secondly, a credibility attenuation factor is defined in the knowledge topology network, and the weights of each node are adjusted based on the user's feedback on different solutions; thirdly, based on the updated node weights, the system uses a path weight redistribution algorithm to determine the optimal retrieval path and its depth from hardware diagnosis to operating system inspection; finally, the system generates multimodal retrieval instructions based on the optimized path weights, drives the parallel extraction module to extract relevant information from text guides, video tutorials and forum discussions, and integrates them into a comprehensive and targeted answer after semantic consistency verification to help users solve problems effectively; through the above steps, not only the accuracy of understanding user needs is improved, but also the system's ability to provide personalized and efficient solutions is enhanced, greatly improving the user experience.

[0049] In order to further improve the flexibility and efficiency of the cache system, in some embodiments, the multi-level cache collaborative optimization is synchronously performed in step 104, the cache semantic coverage and matching priority are dynamically adjusted according to the large model response results and user feedback, and the cache capacity is adaptively expanded and eliminated based on the session traffic characteristics. The logic linkage includes: Build a semantic traceability link, reversely decompose the final response generated by the large model into a contribution distribution map of multi-level cache hit results, and calculate the semantic alignment error of caches at all levels based on the dependency relationship between the nodes in the map; collect user satisfaction scores and follow-up behavior data when the session terminates, establish a cache effect evaluation matrix, and obtain the semantic coverage blind spots and matching rule deviations that need to be corrected for caches at all levels through matrix feature decomposition; analyze the spatiotemporal distribution characteristics of session traffic in real time, build a cache load prediction model based on sliding windows, and dynamically calculate the capacity pressure coefficient of caches at all levels and the hot data migration trend; according to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, synchronously adjust the semantic matching thresholds and storage partitioning strategies of caches at all levels, and trigger the priority reordering and redundant data elimination of cross-layer caches based on the hotspot migration trend.

[0050] This embodiment involves concepts such as semantic traceability link, cache effect evaluation matrix, cache load prediction model, etc. These tools are used to accurately analyze the effectiveness of system response, user satisfaction and cache system performance, so as to dynamically adjust the cache strategy to improve the overall service quality.

[0051] In the embodiment of the present application, firstly, a semantic traceability link is constructed, and the contribution of the multi-level cache is determined by analyzing the response of the large model, and the semantic alignment error is calculated; secondly, user satisfaction scores and follow-up behavior data are collected, a cache effect evaluation matrix is ​​established, and the semantic coverage blind spots and matching rule deviations that need to be improved are identified; thirdly, the sliding window mechanism is used to monitor the session traffic characteristics in real time, predict the cache load, and calculate the pressure coefficient of each level of cache and the hot data migration trend; finally, the semantic matching threshold and storage partition strategy of the cache are synchronously adjusted according to the above analysis results, and the cache priority is rearranged and redundant data is eliminated according to the hot migration trend; Here is a specific example: Suppose that in an intelligent customer service system of an online education platform, a teacher frequently asks questions about the course management function. The system first analyzes the answers to these questions through the semantic traceability link big model, determines the contribution of caches at all levels, and calculates the semantic alignment error; secondly, the system collects the teacher's satisfaction score at the end of the session and whether there is any further questioning behavior, establishes a cache effect evaluation matrix, and finds that the semantic coverage of some specific queries is insufficient; thirdly, the system uses sliding window technology to analyze the time and space distribution characteristics of session traffic, predicts cache load and determines which data is the current hot spot and which can be eliminated; finally, based on the relationship between semantic alignment error and capacity pressure coefficient, the system adjusts the cache semantic matching threshold and storage strategy, and rearranges the cache priority to ensure that the most relevant course management information can be quickly retrieved, while eliminating some outdated or infrequently used data; through the above steps, not only the response speed and accuracy of the cache system are improved, but also the resource allocation is optimized, so that the system can more effectively respond to query needs in high concurrency situations, significantly improving the user experience.

[0052] In order to further improve the adaptability and optimization performance of the cache system, in some embodiments, according to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, the semantic matching threshold and storage partition strategy of each level of cache are synchronously adjusted, and the priority re-arrangement and redundant data elimination of the cross-layer cache are triggered based on the hotspot migration trend, including: A cross-layer association model of the semantic alignment error matrix and the capacity pressure coefficient matrix is ​​constructed, and the collaborative optimization parameters of the multi-level cache are extracted through the feature fusion algorithm to generate a dynamic weight allocation map; based on the dynamic weight allocation map, a path optimization algorithm is used to calculate the semantic matching threshold adjustment amount of each level of cache, and a storage partition strategy update instruction is generated; a priority reordering queue is constructed according to the hotspot migration trend, and a cross-layer cache priority mapping table is generated in combination with the spatiotemporal distribution characteristics of the session traffic; the semantic matching threshold adjustment amount, the storage partition strategy update instruction and the cross-layer cache priority mapping table are input into the cache strategy execution engine, and the threshold update, partition reconstruction and redundant data elimination operations are triggered synchronously, and the adjustment results are fed back to the semantic traceability link for calibration.

[0053] This embodiment involves concepts such as semantic alignment error matrix, capacity pressure coefficient matrix, dynamic weight allocation map, and cross-layer cache priority mapping table. These tools are used to accurately evaluate system performance and dynamically adjust cache strategies accordingly to ensure efficient data management and fast response to user requests.

[0054] In the embodiment of the present application, a cross-layer association model of the semantic alignment error matrix and the capacity pressure coefficient matrix is ​​first constructed, and the collaborative optimization parameters of the multi-level cache are extracted using the feature fusion algorithm to form a dynamic weight allocation map; secondly, the semantic matching threshold adjustment amount of each level of cache is calculated based on the map using the path optimization algorithm, and a new storage partitioning strategy is generated; thirdly, a priority reordering queue is established according to the hotspot migration trend, and a cross-layer cache priority mapping table is generated in combination with the spatiotemporal distribution characteristics of the session traffic; finally, all adjustment information is input into the cache strategy execution engine, and threshold updates, partition reconstructions, and redundant data elimination are performed synchronously, and the final adjustment results are fed back to the semantic traceability link for calibration; Here is a specific example: Assume that in an intelligent customer service system of an e-commerce platform, faced with a large number of queries about product return policies, the system first constructs a cross-layer association model of the semantic alignment error matrix and the capacity pressure coefficient matrix, analyzes the collaborative optimization requirements of the current cache system through the feature fusion algorithm, and generates a dynamic weight allocation map; secondly, based on this map, the system calculates the semantic matching thresholds that need to be adjusted for caches at all levels, and formulates a new storage partitioning strategy; thirdly, the system observes that queries about return policies for specific products have become a hot spot, so a priority reordering queue is established, and a cross-layer cache priority mapping table is generated based on the time and spatial distribution characteristics of session traffic; finally, the system inputs all the above adjustment information into the cache policy execution engine, simultaneously performs threshold updates, partition reconstruction, and redundant data elimination operations, and feeds back these adjustment results to the semantic traceability link for effect verification; through the above steps, not only the flexibility and response speed of the cache system are improved, but also the efficiency of data management is ensured, and the user experience is significantly improved, especially when processing high-concurrency queries.

[0055] In order to further improve the adaptability and optimization performance of the cache system to dynamic traffic, in some embodiments, the real-time analysis of the spatiotemporal distribution characteristics of session traffic, the construction of a cache load prediction model based on a sliding window, and the dynamic calculation of the capacity pressure coefficient of each level of cache and the hot data migration trend include: Based on the sliding window mechanism, the session traffic time series is divided, the interaction frequency, semantic complexity and response delay parameters within the window are extracted, and the spatiotemporal feature vector is generated; the spatiotemporal feature vector is input into the hierarchical prediction model, and the short-term traffic fluctuation characteristics are extracted through the time series convolutional network, and the cross-window traffic evolution law is captured in combination with the long short-term memory network, and the capacity pressure coefficient of each level of cache is output; based on the capacity pressure coefficient, a hotspot data recognition engine is built, and the high-frequency access semantic fragments are screened through the dynamic threshold algorithm, and the hotspot migration path map is generated in combination with the semantic topological relationship; the capacity pressure coefficient and the hotspot migration path map are input into the cache expansion decision module to trigger the dynamic reorganization of the storage partition and the redundant data elimination mechanism, and the load status mark of the semantic traceability link is synchronously updated.

[0056] This embodiment involves concepts such as sliding window mechanism, spatiotemporal feature vector, hierarchical prediction model (including time series convolutional network and long short-term memory network), hot data identification engine and hot migration path map. These tools are used to monitor and analyze the dynamic changes of session traffic in real time to optimize the capacity management and data storage strategy of the cache system.

[0057] In the embodiment of the present application, firstly, a sliding window mechanism is used to process the time series of session traffic, from which parameters such as interaction frequency, semantic complexity and response delay are extracted to form a spatiotemporal feature vector; secondly, these feature vectors are input into a hierarchical prediction model, a temporal convolutional network is used to capture short-term traffic fluctuation characteristics, and a long-term traffic pattern is identified through a long short-term memory network, thereby calculating the capacity pressure coefficient of each level of cache; thirdly, a hotspot data recognition engine is constructed based on the capacity pressure coefficient, a dynamic threshold algorithm is used to screen out frequently accessed semantic fragments, and a hotspot migration path map is generated in combination with the semantic topological relationship; finally, the capacity pressure coefficient and the hotspot migration path map are input into the cache expansion decision module to trigger the dynamic adjustment of storage partitions and the elimination of redundant data, and synchronously update the load status mark in the semantic traceability link; Here is a specific example: Assume that in an intelligent customer service system of an online travel platform, when users frequently query information about popular destinations, the system first uses a sliding window mechanism to analyze the time series of session traffic, extracts the interaction frequency, semantic complexity, and response delay in each window, and generates a spatiotemporal feature vector; secondly, these feature vectors are input into a hierarchical prediction model, and the short-term traffic fluctuation characteristics are identified through a temporal convolutional network. The long short-term memory network then captures the traffic evolution law across windows and calculates the capacity pressure coefficients of caches at all levels; thirdly, the system builds a hot data identification engine based on these coefficients, uses a dynamic threshold algorithm to filter out high-frequency access semantic fragments about specific popular destinations, and generates a hot migration path map; finally, the system inputs the capacity pressure coefficient and the hot migration path map into the cache expansion decision module, triggers the dynamic reorganization of storage partitions and the elimination of redundant data, and simultaneously updates the load status mark in the semantic traceability link; through the above steps, not only the response speed and resource utilization of the cache system are significantly improved, but also the efficiency of data management is ensured, especially when dealing with high-concurrency queries, the color of the multi-level cache mechanism is shown, which greatly improves the user experience.

[0058] Figure 2 A schematic diagram of a large model deployment system is provided for an embodiment of the present application. Figure 2 As shown, the device comprises: Capture module 21 is used to establish a first-level dynamic semantic cache, capture the semantic features of online sessions in real time, encode and generate unique identifiers, count high-frequency semantic fragments based on dynamic access weights, and aggregate discrete semantic fragments to reconstruct complete response logic and give priority feedback when the overlap between the semantic identifier of a new request and the cached fragments meets a preset condition; The identification module 22 is used to start the second-level context association cache for the missed request, extract the implicit semantic trajectory vector of the session, build a dynamic association map in combination with the historical interaction path, identify the potential context dependencies across sessions, and trigger the progressive answer splicing mechanism when it is detected that there are continuous semantic nodes in the association map that form a logical link with the current request; The extraction module 23 is used to activate the third-level intent decision cache when the second-level cache is not covered, classify the request into scenarios and mark the status through a multi-level parsing structure, generate structured query instructions based on the knowledge topology network, dynamically adjust the knowledge retrieval path weight, and extract multimodal response elements to generate a combined response; The adjustment module 24 is used to synchronously perform multi-level cache collaborative optimization, dynamically adjust the cache semantic coverage and matching priority according to the large model response results and user feedback, and realize the adaptive expansion of cache capacity and elimination logic linkage based on session traffic characteristics.

[0059] Figure 2The large model deployment system based on a multi-level cache mechanism can execute Figure 1 The implementation principle and technical effect of the large model deployment method based on the multi-level cache mechanism described in the illustrated embodiment will not be repeated. The specific manner in which each module and unit performs operations in the large model deployment device based on the multi-level cache mechanism in the above embodiment has been described in detail in the embodiment of the method, and the multi-level cache mechanism will not be described in detail here.

[0060] In one possible design, Figure 2 A large model deployment system based on the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0061] The processing component 32 is used for the above Figure 1 The embodiment provides a large model deployment method based on a multi-level cache mechanism.

[0062] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0063] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0064] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0065] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0066] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0067] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0068] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a large model deployment method based on a multi-level cache mechanism.

[0069] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0070] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0071] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A large model deployment method based on a multi-level cache mechanism, characterized in that: The following steps are involved: Establish a first-level dynamic semantic cache to capture the semantic features of online sessions in real time and encode them to generate unique identifiers. Count high-frequency semantic fragments based on dynamic access weights. When the overlap between the semantic identifier of a new request and the cached fragment meets the preset conditions, aggregate the discrete semantic fragments to reconstruct the complete response logic and give priority feedback. For missed requests, the second-level context association cache is started to extract the implicit semantic trajectory vector of the session, and a dynamic association graph is constructed in combination with the historical interaction path to identify potential context dependencies across sessions. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, the progressive answer splicing mechanism is triggered. When the second-level cache is not covered, the third-level intent decision cache is activated, the request is classified into scenarios and marked with status through a multi-level parsing structure, structured query instructions are generated based on the knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and multimodal response elements are extracted to generate a combined response; Synchronously perform multi-level cache collaborative optimization, dynamically adjust cache semantic coverage and matching priority according to large model response results and user feedback, and realize cache capacity adaptive expansion and elimination logic linkage based on session traffic characteristics; Among them, the semantic identifiers of the first-level cache are orthogonally complementary to the semantic trajectory vector space of the second-level cache, the association graph nodes of the second-level cache and the knowledge topology network of the third-level cache are mapped through the semantic bridging layer, and the update conditions of the caches at all levels are dynamically negatively correlated with the complexity of the session flow.

2. The method according to claim 1, characterized in that The second-level context association cache is started for the missed request, the implicit semantic trajectory vector of the session is extracted, a dynamic association graph is constructed in combination with the historical interaction path, and the potential context dependency across sessions is identified. When continuous semantic nodes that form a logical link with the current request are detected in the association graph, a progressive answer splicing mechanism is triggered, including: Extract the context encoding features of the continuous interaction sequence in the current session and generate an implicit semantic trajectory vector with temporal dependencies; The implicit semantic trajectory vector is aligned with the interaction path of the same user identifier or similar user group in the historical session library in time and space, and a dynamic association graph including semantic nodes and relationship edges is constructed, wherein the graph nodes store the semantic context fragments across sessions, and the relationship edges record the logical jump probability and timeliness weight between nodes; Performing bidirectional semantic propagation calculation on the dynamic association graph, activating adjacent nodes on the association path based on the semantic trajectory vector of the current request, and identifying multi-hop semantic chains that form logical coherence with the current session through a path confidence screening algorithm; When it is detected that there are at least two associated nodes with time continuity and intention consistency in the multi-hop semantic chain, the reconstruction engine of the cross-session answer fragments is triggered, and the discrete answer elements are dynamically sorted and connected according to the logical order and weight ratio between the nodes to generate a progressive spliced ​​response that is coherent with the current context.

3. The method according to claim 1, characterized in that The third level intention decision cache is activated when the second level cache is not covered, the request is classified into scenarios and marked with status through a multi-level parsing structure, a structured query instruction is generated based on a knowledge topology network, the knowledge retrieval path weight is dynamically adjusted, and a multi-modal response element is extracted to generate a combined response, including: The session state awareness module captures the intent expression pattern and scenario feature parameters in the current interaction sequence, and generates a multidimensional classification vector containing domain labels and state variables. Mapping the multidimensional classification vector to a pre-constructed knowledge topology network, generating a structured query instruction with path constraints according to the domain association strength of each subgraph in the network, wherein the query instruction includes a search condition priority and a cross-domain fusion rule; Real-time monitoring of focus shifts and feedback signals in user interaction behaviors, dynamic calculation of the credibility attenuation factor of each node in the knowledge retrieval path, and adjustment of the scope and depth of cross-domain retrieval based on the path weight redistribution algorithm; According to the updated retrieval path, multimodal data units such as text, images, and operation instructions are extracted in parallel from heterogeneous data sources. After filtering the conflicting elements through the semantic consistency verification module, the multimodal units are embedded in the dynamically generated dialogue framework according to the preset response assembly template to form a combined response.

4. The method according to claim 1, characterized in that: The synchronous execution of multi-level cache collaborative optimization dynamically adjusts the cache semantic coverage and matching priority according to the large model response results and user feedback, and realizes the adaptive expansion of cache capacity and the linkage of elimination logic based on session traffic characteristics, including: Build a semantic traceability link to reversely decompose the final response generated by the large model into a contribution distribution graph of multi-level cache hit results, and calculate the semantic alignment error of each level of cache based on the dependency relationship between graph nodes; Collect user satisfaction scores and follow-up behavior data when the session ends, establish a cache effect evaluation matrix, and obtain the semantic coverage blind spots and matching rule deviations that need to be corrected for caches at all levels through matrix feature decomposition; Analyze the spatiotemporal distribution characteristics of session traffic in real time, build a cache load prediction model based on sliding windows, and dynamically calculate the capacity pressure coefficient of each level of cache and the migration trend of hot data; According to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, the semantic matching thresholds and storage partitioning strategies of caches at all levels are adjusted synchronously, and priority reordering and redundant data elimination of cross-layer caches are triggered based on hotspot migration trends.

5. The method according to claim 3, characterized in that: The real-time monitoring of the focus shift and feedback signals in the user interaction behavior, the dynamic calculation of the credibility attenuation factor of each node in the knowledge retrieval path, and the adjustment of the scope and depth of the cross-domain retrieval based on the path weight redistribution algorithm include: Based on the user interaction behavior sequence, the attention transfer matrix is ​​constructed, the current interaction focus and historical focus sequence are extracted through the sliding window mechanism, the focus transfer probability is calculated and the focus transfer vector is generated; Define the credibility attenuation factor in the knowledge topology network, calculate the node weight correction value based on user feedback signals, and update the node weight; Based on the updated node weights, a path weight redistribution algorithm is used to calculate the priority and depth limit of the cross-domain retrieval path; A multimodal retrieval instruction is generated according to the path weight coefficient and depth limit, driving the parallel extraction module to extract data units according to the weight ratio and input them into the semantic consistency verification module.

6. The method according to claim 4, characterized in that According to the coupling relationship between the semantic alignment error and the capacity pressure coefficient, the semantic matching thresholds and storage partitioning strategies of caches at all levels are synchronously adjusted, and priority re-arrangement and redundant data elimination of cross-layer caches are triggered based on the hotspot migration trend, including: Construct a cross-layer association model between the semantic alignment error matrix and the capacity pressure coefficient matrix, extract the collaborative optimization parameters of multi-level cache through the feature fusion algorithm, and generate a dynamic weight allocation map; Based on the dynamic weight allocation graph, a path optimization algorithm is used to calculate the semantic matching threshold adjustment amount of each level of cache, and a storage partition strategy update instruction is generated; Build a priority reordering queue based on the hotspot migration trend, and generate a cross-layer cache priority mapping table based on the spatiotemporal distribution characteristics of session traffic; The semantic matching threshold adjustment amount, the storage partition strategy update instruction and the cross-layer cache priority mapping table are input into the cache strategy execution engine, and the threshold update, partition reconstruction and redundant data elimination operations are triggered synchronously, and the adjustment results are fed back to the semantic traceability link for calibration.

7. The method according to claim 4, characterized in that The real-time analysis of the spatiotemporal distribution characteristics of session traffic, the construction of a cache load prediction model based on a sliding window, and the dynamic calculation of the capacity pressure coefficient of each level of cache and the migration trend of hot data include: Based on the sliding window mechanism, the session traffic time series is divided, the interaction frequency, semantic complexity and response delay parameters within the window are extracted, and the spatiotemporal feature vector is generated; The spatiotemporal feature vector is input into the hierarchical prediction model, the short-term traffic fluctuation characteristics are extracted through the time series convolution network, the cross-window traffic evolution law is captured by combining the long short-term memory network, and the capacity pressure coefficient of each level of cache is output; Based on the capacity pressure coefficient, a hotspot data identification engine is constructed, high-frequency access semantic fragments are screened through a dynamic threshold algorithm, and a hotspot migration path map is generated in combination with semantic topological relationships; The capacity pressure coefficient and the hotspot migration path map are input into the cache expansion decision module to trigger the dynamic reorganization of storage partitions and the redundant data elimination mechanism, and the load status mark of the semantic traceability link is synchronously updated.

8. A large model deployment system based on a multi-level cache mechanism, characterized in that: include: The capture module is used to establish the first-level dynamic semantic cache, capture the semantic features of online sessions in real time, encode and generate unique identifiers, count high-frequency semantic fragments based on dynamic access weights, and aggregate discrete semantic fragments to reconstruct the complete response logic and give priority feedback when the overlap between the semantic identifier of the new request and the cached fragment meets the preset conditions; The recognition module is used to start the second-level context association cache for missed requests, extract the implicit semantic trajectory vector of the session, build a dynamic association graph based on the historical interaction path, identify the potential context dependencies across sessions, and trigger the progressive answer splicing mechanism when continuous semantic nodes that form a logical link with the current request are detected in the association graph; The extraction module is used to activate the third-level intent decision cache when the second-level cache is not covered, classify the request scenarios and mark the status through a multi-level parsing structure, generate structured query instructions based on the knowledge topology network, dynamically adjust the knowledge retrieval path weight, and extract multimodal response elements to generate a combined response; The adjustment module is used to synchronously perform multi-level cache collaborative optimization, dynamically adjust the cache semantic coverage and matching priority according to the large model response results and user feedback, and realize the adaptive expansion of cache capacity and elimination logic linkage based on session traffic characteristics.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a large model deployment method based on a multi-level cache mechanism as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a large model deployment method based on a multi-level cache mechanism as described in any one of claims 1 to 7 is implemented.

Images